Académique Documents
Professionnel Documents
Culture Documents
I. INTRODUCTION
LL-OPTICAL networks promise large bandwidth and information carrying capacity [1]. These networks consist of
optical fibers for transporting information over long distances.
Optical switches are used for manipulating the optical signals
directly rather than converting them into the electrical domain
for manipulation, thereby reducing power consumption and increasing network speed and data integrity. All-optical data manipulation is enabled by optical integrated circuits (OICs), or
integrated optics [2].
Manuscript received July 3, 2004; revised March 6, 2005. This work was supported by a National Science Foundation (NSF) CAREER award (R. Ghodssi)
and by the Laboratory for Physical Sciences (LPS). Subject Editor O. Solgaard.
M. W. Pruessner, M. Datta, D. P. Kelly, and R. Ghodssi are with the MEMS
Sensors and Actuators Lab, The Institute for Systems Research, University of
Maryland, College Park, MD 20742 USA. They are also with the Department
of Electrical and Computer Engineering, University of Maryland, College,
Park, MD 20742 USA and the Laboratory for Physical Sciences, College Park,
MD 20740 USA (e-mail: marcelp@glue.umd.edu; mdatta@glue.umd.edu;
dpkelly@glue.umd.edu; ghodssi@eng.umd.edu).
K. Amarnath and P.-T. Ho are with the Department of Electrical and Computer Engineering, University of Maryland, College, Park, MD 20742 USA.
They are also with the Laboratory for Physical Sciences, College Park, MD
20740 USA (e-mail: akuldeep@glue.umd.edu; ho@eng.umd.edu).
S. Kanakaraju is with the Department of Electrical and Computer Engineering, University of Maryland, College, Park, MD 20742 USA (e-mail:
sk@lps.umd.edu).
Digital Object Identifier 10.1109/JMEMS.2005.851848
1071
Fig. 1. (a) Top view schematic (in actuated state with waveguides pulled-in),
(b) optical coupling via interference of odd and even modes, and (c) variation of
coupled optical power in the BAR and CROSS outputs as a function of position
along the length of the waveguide. Note that (b) represents the electric fields;
j j .
the optical power in (c) is
P= A
1072
n = 5 2 10 cm
(
0 05%)
= 3 195 In Ga As P
3 173 (In Ga As P )
n 10 cm
B. Optical Design
The MEMS evanescent coupler utilizes our suspended waveguide technology [35]. Here, the waveguides are suspended
wide tethers spaced
in air above the substrate by 1.5
apart. Optical coupling in our device depends
10002000
,
on the waveguide separation, the physical coupling length,
and the optical polarization. We obtain the coupling coefficient,
, by simulating [36] the effective refractive index for even and
and
) for the 2
wide waveguides
odd modes (
and subsequently calculating the coupling coefficient, , and
[21], [22]:
(4)
(5)
where
and
is the free-space wavelength. Here,
is the characteristic length that results in 100%
coupling from the BAR to the CROSS waveguide; so, ideally,
.
we want
, versus
The simulated characteristic coupling length,
waveguide separation is shown in Fig. 3(a), along with the corresponding coupling coefficient, . Looking at the simulation
result, we note that S-polarization (TE-modes, electric field
perpendicular to wafer plane) results in significantly shorter
than P-polarization (TM-modes, electric field parallel to wafer
plane) for our waveguides. We expect eight times as efficient
coupling for S-polarization compared to P-polarization [see
Fig. 3(b)]. Compact devices with coupling lengths of only a
few hundred microns are feasible. Also, changing the gap from
100 nm to 600 nm effectively turns the device from completely
ON (100% coupling) to OFF (negligible coupling). This
1073
TABLE I
CALCULATED PULL-IN VOLTAGE FOR VARIOUS COUPLER DESIGNS
Fig. 4. (a) Top view of fabricated coupler switch, (b) SEM of suspended 2 m
wide waveguides, and (c) cleaved test waveguide before sacrificial etch with
detail of sidewall roughness. This particular device (coupler 1a) has a 1 m
nominal gap before actuation [(a), (b)]. Other devices have a 2 m nominal
gap.
1074
Fig. 5. Simulated (lines) and measured (symbols) pull-in voltage for various
coupler test structures. Inset: test structure for measuring pull-in voltage.
1075
Fig. 6. Measured coupling length for electrostatically actuated coupler 1a during pull-in. The measured coupling length is L
has a small but visible gap separating the waveguides during pull-in.
Fig. 7. (a) Coupler 2b after actuation and subsequent stiction, (b) detail of center region (note: image height has been stretched in order to show detail of stiction
region). With V = 0 the coupled region is small (solid circle) and with V > 0 the coupling region increases (dashed circle).
Fig. 8. Optical response for coupler 2b: (a) actuation signal, (b) measured
optical BAR output, (c) measured optical CROSS output, and (d) coupling
+P
). Both P
and P
are
loss, defined as loss = 1 (P
normalized to P
.
The measured switching speed of three electrostatically actuated couplers (coupler 1a, coupler 2a, and coupler 2b) is shown
in Fig. 9. The risetime [Fig. 9(a), (c), and (e)] exhibits a delay
before optical coupling occurs due to the waveguide travel time,
the time required for the waveguides to come into sufficiently
close contact in order for optical coupling to occur. It is interesting to note that coupler 1a and coupler 2a exhibit a clear
gap [see Fig. 9(a) and
delay (10 , 18 ) due to the 12
resulting from
(c)]. Coupler 2b has a fairly small delay
the close proximity of the waveguides due to stiction [Fig. 9(e)].
This is in good agreement with theory: due to the exponential
dependence of the coupling coefficient on the waveguide separation, only small gaps ( 100 nm) result in measurable optical
coupling. For this same reason, the falltime [see Fig. 9(b), (d),
1076
TABLE II
SUMMARY OF MEASURED P
AND PHYSICAL COUPLING LENGTH, L , FOR DIFFERENT DEVICES AND ACTUATION VOLTAGES. THE EXTRACTED COUPLING
, AND THE CHARACTERISTIC COUPLING LENGTH, L , FOR S-POLARIZATION ARE ALSO GIVEN
COEFFICIENT,
and (f)] does not show a delay since the beams separate immediately upon removal of the actuation signal and optical coupling decreases. The fastest switching speed we obtained was 4
(coupler 2b). Concerning switching frequency, we actuated
our devices up to 25 kHz (coupler 1a) with no change in device
performance.
1077
Fig. 10. (a) Variable coupling for coupler 2b and (b) variable coupling for coupler 3b. For coupler 2b an actuation voltage of V = 6 Vp
while for coupler 3b an actuation voltage of V = 5 Vp p causes a decrease in P
.
in P
Fig. 10(b)]. At
we turn the device completely OFF
waveguide. By inso that all the optical power is in the
creasing the actuation voltage to 10 V we increase the
power and decrease the
power arriving again at a 3 dB
coupler. This clearly demonstrates the mode interference picture
predicted by (1), (2) and Fig. 1(c).
While the above experiments utilized couplers and stiction,
we note that similar experiments can be made with couplers
that do not rely on pull-in actuation. From Table II we see
, results in larger coupled
that increased coupling length,
power. Alternatively, (2) tells us that an increased coupling
coefficient, , can give us increased coupled power, since
. Smaller waveguide gaps result in
larger coupling coefficient, . Therefore, variable optical
coupling can also be achieved using actuators that enable continuous variation of the waveguide gap, such as comb-drives. To
this end, a MEMS variable coupler with comb-drive actuation
is presented in Section V.
G. Polarization and Wavelength Dependence
Polarization-dependent coupling was measured using coupler 2b (see Fig. 7) during actuation. The measured
coupled power for both S-and P-polarizations during actuation
is shown in Fig. 11. The results indicate 66% coupling for S-and
2.8% coupling for P-polarization. From the S-polarization mea. Similarly, we obtain
surement we obtain
for P-polarization. The polarization-depen, which is in
dent coupling ratio is then
general agreement with the simulated value
[see Fig. 3(b)].
Some of the discrepancy between simulation and experiment
is due to measurement error, taking into account scattered light
in our measurement setup. We note that the coupled power for
levels) so that meaP-polarization is small (2.8%, or subsurement errors are increased compared to S-polarization. Furthermore, some polarization conversion is to be expected in
the waveguides due to sidewall roughness [42] and the trapezoidal waveguide cross-section. Therefore, if we input 100%
input, then some portion will
P-polarized light into the
be converted to S-polarized light resulting in increased coupling compared to 100% P-polarized light. A third factor is that
the waveguides are not strictly single-mode so that higher order
0 p causes an increase
1078
V. DISCUSSION
A. Remarks
Coupled-mode theory [21], [22] is accurate for waveguides
that are weakly coupled. We chose our waveguide dimensions
core cross section) for mechanical robustness
based on our prior work on similar dimensioned structures [14].
However, this results in a waveguide that supports more than one
optical mode. Furthermore, one might argue that during pull-in
the waveguides are in contact and therefore strongly-coupled so
that the coupled-mode theory no longer applies.
The experimental results indicate that our devices can
be modeled by single-mode waveguides. Looking at the
single-mode simulations (Fig. 3), we see that for 100200 nm
waveguide gaps we obtain characteristic coupling lengths
. This length scale agrees well with the
, for coupler 1a and
measured pull-in coupling lengths,
coupler 2b (Table II). For coupler 3b we initially obtain lower
coupling with increasing voltage due to destructive interference, in agreement with the coupled-mode theory assuming
single-mode waveguides.
If higher-order modes contribute significantly to coupling,
we should see much stronger coupling than what is predicted
by simulation (see Fig. 3). This stronger coupling should result
.
in significantly shorter characteristic coupling lengths,
However, the experimental results are in good agreement with
simulation based on the fundamental mode, both in the coupling lengths required as well as the polarization-dependence.
Furthermore, due to our suspended waveguide design in which
we utilize tethers for support [35], higher order modes will
experience increased losses compared to the first-order mode.
The reason for this increased loss is the tighter confinement
of the fundamental mode to the core of the waveguide, while
higher order modes are less confined. Therefore, the fundamental mode dominates optical coupling from the BAR to the
CROSS waveguide.
In order to obtain more efficient coupling with greater than
66% coupled power to the CROSS waveguide, we can change
either 1) the coupling coefficient, , or 2) the physical coupling length,
. From Table II we see that the coupling coefficient, , is relatively constant for different devices and actuation voltages. From and (5) we obtain the characteristic cou, which determines the 100% coupling length.
pling length,
Looking at Table II, we see that the physical coupling length is
well below the characteristic coupling length, so that
for coupler 1a and coupler 2b. In order to increase coupling in our devices to 100% we therefore need to increase the
. Alternatively, by designing narpull-in length to
) the evanescent
rower single-mode waveguides (width
field is increased, resulting in a larger coupling coefficient, ,
and increased coupled power.
1079
Fig. 12. Schematics for 1 2 couplers: (a) comb-drive coupler in rest state, (b) comb-drive coupler in actuated state, (c) third-electrode coupler in rest state, and
(d) third-electrode coupler in actuated state.
of the MEMS coupler is excellent ( 47 dB) due to the exponential dependence of the coupling coefficient on waveguide gap.
This is a significant advantage of our MEMS approach, since
small actuation distances result in large changes in coupling.
Finally, while the coupling efficiency in the present devices
can be improved compared to other more mature coupler
approaches, single-mode waveguides with matched coupling
) should result in complete coupling in
lengths (i.e.,
our MEMS switch.
VI. CONCLUSION
Fig. 13. Optical coupling in coupler C-D [1
Fig. 12(a) and (b)].
to
waveguide separation, then we can avoid pull-in
since pull-in is known to occur after a travel range of one third
the original gap [46]. This enables continuous variation of the
gap and hence enables variable optical coupling.
A third approach for increasing device reliability makes use
of surface coatings. Self-assembled monolayers [47] can be utilized to increase the water contact angle of InP surfaces, thereby
making InP hydrophobic. This has the potential for preventing
stiction in our pull-in type MEMS couplers.
C. Comparison of Device Performance
The MEMS coupler experiments show excellent performance compared with other competing designs, such as electrothermal and electro-optic coupler switches. We demonstrated
low-voltage/low-power electrostatic actuation. Electrothermal
of
couplers typically consume mW power [23], so even the
pull-in current in our MEMS couplers results in significantly
lower power consumption. Fabrication of our devices is simple
since the waveguides and actuation elements are fabricated
using standard optical lithography. Although we use InP, the
MEMS coupler can be implemented in low-cost silicon-based
materials and does not rely on specific material properties (such
switching speed is
as electro-optic effects [24]). The 420
significantly faster than that of electrothermal couplers [23] as
well as other MEMS optical switches [16][19], although it
does not compare with the ns (or even sub-ns) speed obtainable
)
with electro-optic switching. The channel isolation (at
1080
REFERENCES
[1] C. A. Brackett, Dense wavelength division multiplexing networks:
principles and applications, IEEE J. Sel. Areas Commun., vol. 8, pp.
948964, 1990.
[2] E. Pennings, G. Khoe, M. K. Smit, and T. Staring, Integrated-optic
versus microoptic devices for fiber-optic telecommunication systems:
a comparison, IEEE J. Sel. Topics Quantum Electron., vol. 2, pp.
151164, 1996.
[3] H. Fujita and H. Toshiyoshi, Optical MEMS, IEICE Trans. Elec., vol.
E83, pp. 14271434, 2000.
[4] J. A. Walker, The future of MEMS in telecommunications networks,
J. Micromech. Microeng., vol. 10, pp. R1R7, 2000.
[5] S. Adachi, Physical Properties of III-V Semiconductor Compounds. New York: Wiley, 1992.
[6] S. Greek, R. Gupta, and K. Hjort, Mechanical considerations in the
design of a micromechanical tuneable InP-based WDM filter, J. Microelectromech. Syst., vol. 8, pp. 328334, 1999.
[7] S. Irmer, J. Daleiden, V. Rangelov, C. Prott, F. Rmer, M. Strassner, A.
Tarraf, and H. Hillmer, Ultralow biased widely continuously tunable
Fabry-Prot filter, IEEE Photon. Technol. Lett., vol. 15, pp. 434436,
2003.
[8] J. L. Leclercq, M. Garrigues, X. Letartre, C. Seassal, and P. Viktorovitch,
InP-based MOEMS and related topics, J. Micromech. Microeng., vol.
10, pp. 287292, 2000.
[9] R. Ledantec, T. Benyattou, G. Guillot, C. Seassal, J. L. Leclercq,
X. Letartre, A. Gagnaire, M. Gendry, P. Viktorovitch, R. Benferhat,
D. Rondi, and R. Blondeau, Optical characterization methods of
InP-based micro-opto-electro-mechanical systems, SPIE, vol. 3008,
pp. 258264, 1997.
[10] R. Ledantec, T. Benyattou, G. Guillot, A. Spisser, C. Seassal, J. L.
Leclercq, P. Viktorovitch, D. Rondi, and R. Blondeau, Tunable microcavity based on InP-air Bragg mirrors, IEEE J. Sel. Topics Quantum
Electron., vol. 5, pp. 111114, 1999.
[11] A. Spisser, R. Ledantec, C. Seassal, J.-L. Leclercq, T. Benyattou, D.
Rondi, R. Blondeau, G. Guillot, and P. Viktorovitch, Highly selective
and widely tunable 1.55 um InP/air-gap micromachined Fabry-Perot
filter for optical communications, IEEE Photon. Technol. Lett., vol. 10,
pp. 12591261, 1998.
[12] M. Strassner, J. Daleiden, N. Chitica, D. Keiper, D. Stalnacke, S. Greek,
and K. Hjort, III-V semiconductor material for tunable Fabry-Perot filters for coarse and dense WDM systems, Sensors and Actuators A, vol.
A85, pp. 249255, 2000.
[13] S. Greek, K. Hjort, J.-A. Schweitz, C. Seassal, J. L. Leclercq, M. Gendry,
M. P. Besland, P. Viktorovitch, C. Figuet, V. Souliere, and Y. Monteil,
The strength of indium phosphide based microstructures, SPIE, vol.
3008, pp. 251257, 1997.
[14] M. W. Pruessner, T. King, D. Kelly, R. Grover, L. C. Calhoun, and
R. Ghodssi, Mechanical property measurement of InP-based MEMS
for optical communications, Sens. Actuators A, Phys., vol. A105, pp.
190200, 2003.
[15] E. Gini and H. Melchior, Thermal dependence of the refractive index
of InP measured with integrated optical demultiplexer, J. Appl. Phys.,
vol. 79, pp. 43354337, 1996.
[16] E. Ollier, P. Labeye, and F. Revol, Micro-opto mechanical switch integrated on silicon, Electron. Lett., vol. 31, pp. 20032005, 1995.
[17] E. Ollier, Optical MEMS devices based on moving waveguides, IEEE
J. Sel. Topics Quantum Electron., vol. 8, pp. 155162, 2002.
[18] T. Bakke, C. P. Tigges, J. J. Lean, C. T. Sullivan, and O. B. Spahn,
Planar microoptomechanical waveguide switches, IEEE J. Sel. Topics
Quantum Electron., vol. 8, pp. 6472, 2002.
[19] T. Bakke, C. P. Tigges, and C. T. Sullivan, 1 2 MOEMS switch based
on silicon-on-insulator and polymeric waveguides, Electronics Letters,
vol. 38, pp. 177178, 2002.
[20] G. Keiser, Optical Fiber Communications, 3rd ed. Boston, MA: McGraw Hill, 2000.
[21] K. Okamoto, Fundamentals of Optical Waveguides. San Diego, CA:
Academic, 2000.
[22] R. G. Hunsperger, Integrated Optics: Theory and Technology, 5th
ed. Berlin, Germany: Springer, 2002.
[23] Q. Lai, W. Hunziker, and H. Melchior, Low-power compact 2 2 thermooptic silica-on-silicon waveguide switch with fast response, IEEE
Photon. Technol. Lett., vol. 10, pp. 681683, 1998.
[24] S. A. Samson, R. F. Tavlykaev, and R. V. Ramaswamy, Two-section
reversed switch with uniform electrodes and domain reversal, IEEE
Photon. Technol. Lett., vol. 9, pp. 197199, 1997.
Madhumita Datta (S98M02) received the M.S. and Ph.D. degrees in electrical and computer engineering, both from the University of Maryland, College Park, in 1999 and 2002, respectively. Prior to that, she received the B.S.
degree in electrical engineering from Jadavpur University, Calcutta, India, in
1995. Her doctoral research was on hybrid integration and packaging of optoelectronic and fiber-optic modules for telecom and datacom applications.
As part of her postdoctoral research at the MEMS Sensors and Actuators Laboratory at the University of Maryland, College Park, she works on
MEMS-tunable wavelength-selective integrated photonic devices for reconfigurable wavelength division multiplexing (WDM) networks and BioMEMS
applications. She has published a number of peer-reviewed articles on optoelectronic packaging and optical MEMS in various international journals and
conference proceedings, and holds one U.S. patent.
1081